igraph

• A graph in graph theory is defined by vertices and edges. The vertices are connected by edges.

• My preference was to start with an empty graph and then add the vertices.

library("igraph")
mygraph <- make_empty_graph(n = 0, directed = FALSE)
mygraph <- mygraph + vertices(c("Vertex A", "Vertex B"), type = "skill")
V(mygraph)$size <- 40.0
E(mygraph)$width <- 30.0

plot(mygraph,  vertex.label.cex=10, vertex.label.dist = 2.5, vertex.label.degree=pi)

• and then add the edges…

mygraph <- mygraph + edge("Vertex A", "Vertex B")
V(mygraph)$size <- 40.0
E(mygraph)$width <- 30.0

plot(mygraph,  vertex.label.cex=10, vertex.label.dist = 2.5, vertex.label.degree=pi)

• and then add a vertex and some more edges

mygraph <- mygraph + vertex("Vertex C", type = "company")
mygraph <- mygraph + edge("Vertex A", "Vertex C")
V(mygraph)$size <- 40.0
E(mygraph)$width <- 30.0

plot(mygraph,  vertex.label.cex=10, vertex.label.dist = 2.5, vertex.label.degree=pi)

• Analytics project with a member of the Human Resources team
• Objective was to see how companies and skills were related in a specific market
• Ideally the output was to look like an old Linkedin inMap which is no longer supported.

• The network was developed by making vertices with each of the candidates skills and edges between them

mygraph <- make_empty_graph(n = 0, directed = FALSE)
item_skills <- c("Skill A", "Skill F", "Skill C", "Skill H")
item_company <- c("Company A")
mygraph <- mygraph + vertices(item_skills, type = "skill")
if (length(item_skills)>1) {
 for (j in 1:(length(item_skills)-1)) {
   for (k in (j+1):length(item_skills)) {
     mygraph <- mygraph + edge(item_skills[j],item_skills[k])   } } }
V(mygraph)$size <- 40.0
E(mygraph)$width <- 30.0

plot(mygraph,  vertex.label.cex=10, vertex.label.dist = 2.5, vertex.label.degree=pi)

• Each of the skills are connected to the candidate's company

if (length(item_skills)>0) {
     mygraph <- mygraph + vertex(item_company[1], type = "company")
     for (j in 1:length(item_skills)) {
       mygraph <- mygraph + edge(item_skills[j], item_company[1])     }   }
V(mygraph)$size <- 40.0
E(mygraph)$width <- 30.0

plot(mygraph,  vertex.label.cex=10, vertex.label.dist = 2.5, vertex.label.degree=pi)

• Do it again for each of the candidates

item_skills2 <- c("Skill A", "Skill G", "Skill C", "Skill J")
item_company2 <- c("Company C")
mygraph <- mygraph + vertices(setdiff(item_skills2,item_skills), type = "skill")
if (length(item_skills2)>1) {
 for (j in 1:(length(item_skills2)-1)) {
   for (k in (j+1):length(item_skills2)) {
     mygraph <- mygraph + edge(item_skills2[j],item_skills2[k])   } } }
V(mygraph)$size <- 40.0
E(mygraph)$width <- 30.0

plot(mygraph,  vertex.label.cex=10, vertex.label.dist = 2.5, vertex.label.degree=pi)

• I didn't code it like this but this is the way I added candidate information

if (length(item_skills2)>0) {
     mygraph <- mygraph + vertex(item_company2[1], type = "company")
     for (j in 1:length(item_skills2)) {
       mygraph <- mygraph + edge(item_skills2[j], item_company2[1])     }   }
V(mygraph)$size <- 40.0
E(mygraph)$width <- 30.0
V(mygraph)$color[which(V(mygraph)$type == "skill")] <- "light blue"
V(mygraph)$color[which(V(mygraph)$type == "company")] <- "gold"

plot(mygraph,  vertex.label.cex=10, vertex.label.dist = 2.5, vertex.label.degree=pi)

• This was the graph after the edges were consolidated and a corresponding edge weight was computed
• The graph uses the Fruchterman-Reingold layout algorithm with edge weights
• There are 263 vertices, well over the 100 vertices that considered too many to display, aka hairball
• It was clear enough that we could make out potential communities

• This is the graph with the spin-glass model communities colored in
• The vertices are in the same location just the colors were changed

• We were asked to generate graph ranking metrics